Group Graduation and Conference Photos by Face

Event photos often show the same person across arrivals, group shots, stage appearances and informal moments. Face grouping gathers likely matches for review. It does not determine a person's identity or automatically assign real names.

1. Organize the event folders

Use an event or class folder, with subfolders for photographers, cameras or sessions. This makes it easier to limit clustering to the event and keep unrelated projects out of the results.

2. Enable face parsing during import

Add the event folder, include subfolders when needed and enable Face Search. If the job only needs face grouping, you can omit unrelated image-search strategies. The separate Find Duplicates switch is optional for later duplicate checks. The screenshot has no folder selected, so Start Adding is disabled. Choose the event folder before starting, and test a small batch before processing a very large event.

3. Wait for the face tasks

Open the task indicator and wait for detection and feature extraction to finish. Partial import results can produce incomplete groups. Large group photos may take longer because each contains many faces.

4. Investigate missed small faces

For large group photos with small faces, review Settings → Face and its large-image tiling parameters. Change them gradually, restart as required and reprocess the relevant face data. Ordinary close portraits usually do not need this adjustment.

5. Open Face Album and select the event

Use More → Face Clustering in the workspace. After model initialization, Face Album opens. Select the event in Folder Filter before summarizing.

Without a folder restriction, clustering can include the whole database. Use that for long-term library organization, not when delivering only one event.

6. Choose a threshold and summarize

The screenshots illustrate the steps using one public portrait and its identical backup: threshold 40 produces one group with two photos. They do not demonstrate recognition accuracy across different event portraits, and 40 is not a universal recommended threshold. Review real event photos as described below.

A lower similarity threshold groups more loosely and can merge similar-looking people. A higher threshold is stricter but can split one person's profiles, expressions or obscured faces across groups.

Start with the default. If different people are merged, increase it; if one person is split too widely, try reducing it and review again.

7. Review the matching faces

The number at the top left of a card is the group ID; the photo-count badge is at the top right. Open each group and inspect its matching faces. Covers and group thumbnails focus on the relevant face even in a group photograph, independently of the general face-zoom setting. Open the complete image when necessary.

Profiles, masks, glasses, makeup, backlighting, low resolution and similar-looking people can cause mistakes. Review important recipients and uncertain groups before delivery.

8. Export group folders

The top Export button includes all current groups, not just a selected card. To deliver one event, restrict the source with Folder Filter and summarize again. After reviewing the grouping, choose Export, check the group count in the confirmation and select the destination. Wait for export to complete.

Folders use group numbers, not automatically recognized real names. Rename only after confirming the person; keep uncertain groups numbered. Source access and destination space still matter.

9. Search for missing photos

Use Face Search with a clear reference, select the target face and restrict the search to the event folder. This is useful when a parent, organizer or colleague asks for additional pictures of one person.

Practical organization and troubleshooting

For graduation work, group by class or event date. For conferences, separate sessions or venues and review key speakers first. For company events, use department or activity folders where useful.

• Empty album: complete face parsing for the selected library and scope.
• Different people in one group: raise the threshold and review similar-looking faces.
• One person split across groups: lower the threshold or merge verified exported folders manually.
• Missed small faces: adjust the relevant face settings and reprocess a sample first.
• Need maintained names and corrections: use the People panel for naming, merging, ignoring and correcting assignments.

Photo counts, face counts and RAW export

Photo counts and face counts can differ. A photo may contain several relevant face records, and exported-photo totals need not equal face totals.

RAW export copies original RAW files, not preview images. Ensure the originals are accessible and the destination has enough capacity. Face similarity remains an organizational clue, so inspect full images before handing over the folders.

Sample image: NASA public-domain astronaut image from the scikit-image sample dataset. The export screenshot shows the confirmation scope; copying still requires choosing a destination and waiting for completion.